重构MQL5中的经典策略（第三部分）：富时100指数预测·综合运用
「用 CTrade 平掉 UK100 持仓」
在 MQL5 里用标准库交易，最省事的就是 CTrade 类的 PositionClose。传入品种名即可对当前账户下该品种的所有持仓下达市价平仓指令。 下面这行就是直接平掉 UK100 的写法，常见于定时器或信号触发的回调里。它不关心开仓价格，只按当时市场价成交，滑点风险随流动性变化。 外汇与指数差价合约杠杆高，UK100 这类指数品种跳空频繁，实盘前务必在 MT5 策略测试器用历史数据跑一遍确认逻辑。
Trade.PositionClose("UK100");◍ 用马科维茨思路压低组合方差
前面那版交易应用波动太跳，根源在单标的权重没约束。借用马科维茨的现代投资组合理论，把注意力从「哪只票能跑赢」转到「整体方差最小」，思路是少碰高方差标的、多配低方差标的，用解析法替掉拍脑袋。 先把富时100股票池的收益率拉出来（用收益率而非收盘价，且每个值乘100得到百分比收益）。池子里像 Ashtead Group（AHT.LSE）尾部明显偏离11只票的平均表现，这类高离散度标的就是优化时要削减的对象。直接算相关性热力图没看到显著相关，即便做20根K线的领先-滞后处理（股票池后移20、UK100前移20）也无收获。 真正动手的是方差最小化：11个权重对应11只票，系数正负代表买卖，区间锁在[-1,1]，且L1范数必须等于1（即权重绝对值之和为1，保证全部资本派出）。初始随机权重算出的组合方差是 0.011959689589562724，成本函数就是「当前权重下的组合方差」，用线性代数点乘即可。 用 SciPy 的 SLSQP（迪特尔·克拉夫特80年代拟牛顿法，估海森矩阵）跑优化，结果 fun 降到 0.0004706570068070814，L1范数校验为 0.9999999998893961（浮点误差所致）。按最优权重换算手数：每开1手 UK100 多头，ADM.LSE 不动、AAL.LSE 开2手等量空头、AV.LSE 开4手空头、BP.LSE 与 BKG.LSE 各开1手多头，其余不动。外汇/贵金属自带高杠杆高风险，这套权重逻辑在 MT5 实盘前务必用历史数据复算。
class="macro">#Import the libraries we need class="kw">import pandas as pd class="kw">import numpy as np class="kw">import seaborn as sns class="kw">import MetaTrader5 as mt5 class="kw">import matplotlib.pyplot as plt from scipy.optimize class="kw">import minimize class="macro">#Create the list of stocks stocks = ["ADM.LSE","AAL.LSE","ANTO.LSE","AHT.LSE","AZN.LSE","ABF.LSE","AV.LSE","BARC.LSE","BP.LSE","BKG.LSE","UK100"] class="macro">#Initialize the terminal if(!mt5.initialize()): print(&class="macro">#x27;Failed to load the MT5 Terminal&class="macro">#x27;) class="macro">#Create a dataframe to store our returns amount = class="num">10000 returns = pd.DataFrame(columns=stocks,index=np.arange(class="num">0,amount)) class="macro">#Fetch the data for stock in stocks: temp = pd.DataFrame(mt5.copy_rates_from_pos(stock,mt5.TIMEFRAME_M1,class="num">0,amount)) returns[[stock]] = temp[[&class="macro">#x27;close&class="macro">#x27;]] class="macro">#Store the data as returns returns = returns.pct_change() returns.dropna(inplace=True) class="macro">#Let&class="macro">#x27;s look at our dataframe returns = returns * (class="num">10.0 ** class="num">2) class="macro">#Let&class="macro">#x27;s visualize our market returns returns.plot() class="macro">#Let&class="macro">#x27;s analyze the correlation coefficients fig, ax = plt.subplots(figsize=(class="num">8,class="num">8)) sns.heatmap(returns.corr(),annot=True,linewidths=.class="num">5, ax=ax) # Let&class="macro">#x27;s also analyze for lead-lag correlation look_ahead = class="num">20 lead_lag = pd.DataFrame(columns=stocks,index=np.arange(class="num">0,returns.shape[class="num">0] - look_ahead)) for stock in stocks: if stock == &class="macro">#x27;UK100&class="macro">#x27;: lead_lag[[stock]] = returns[[stock]].shift(-class="num">20) else: lead_lag[[stock]] = returns[[stock]].shift(class="num">20) # Returns lead_lag.dropna(inplace=True) class="macro">#Let&class="macro">#x27;s see if there are any stocks that are correlated with the future UK100 returns fig, ax = plt.subplots(figsize=(class="num">8,class="num">8)) sns.heatmap(lead_lag.corr(),annot=True,linewidths=.class="num">5, ax=ax) class="macro">#Let&class="macro">#x27;s attempt to minimize the variance of the portfolio weights = np.array([class="num">0,class="num">0,class="num">0,class="num">0,class="num">0,class="num">0,-class="num">1,class="num">1,class="num">1,class="num">0,class="num">0]) covariance = returns.cov() class="macro">#Store the initial portfolio variance initial_portfolio_variance = np.dot(weights.T,np.dot(covariance,weights)) initial_portfolio_variance class="macro">#Cost function def cost_function(x): class="kw">return(np.dot(x.T,np.dot(covariance,x))) class="macro">#Constraints def l1_norm(x):
用 SLSQP 把 11 资产组合方差压到最小
这段脚本在做的，是把一个 11 维权重向量丢进 SciPy 的 SLSQP 求解器，在 L1 范数等于 1 的约束下最小化组合方差。权重边界设在 [-1, 1]，意味着允许反向持仓，对贵金属与外汇交叉盘的对冲组合有现实意义。 约束函数写的是 return(np.sum(np.abs(x))) - 1，即强制绝对值权重之和为 1，而非简单加权和为 1。这样求解出来的 optimal_weights 经 np.sum(np.abs(optimal_weights)) 验证后仍满足该条件。 新旧方差被同时乘以 1e7 后塞进 DataFrame 画柱状图，方便直观对比压缩效果。最后一行 int_weights = (optimal_weights / optimal_weights[-1]) // 1 是把权重按末资产归一并取整，得到可手搓的 integer 手数比例。外汇与贵金属杠杆高，这类优化仅降低方差、不消除爆仓风险，实盘前请在 MT5 策略测试器用历史数据复核。
class="kw">return(np.sum(np.abs(x))) - class="num">1 constraints = {&class="macro">#x27;type&class="macro">#x27;: &class="macro">#x27;eq&class="macro">#x27;, &class="macro">#x27;fun&class="macro">#x27;:l1_norm} class="macro">#Initial guess initial_guess = weights class="macro">#Add bounds bounds = [(-class="num">1,class="num">1)] * class="num">11 class="macro">#Minimize the portfolio variance result = minimize(cost_function,initial_guess,method="SLSQP",constraints=constraints,bounds=bounds) class="macro">#Store the optimal weights optimal_weights = result.x class="macro">#Validating the weights add up to one np.sum(np.abs(optimal_weights)) class="macro">#Store the new portfolio variance otpimal_variance = cost_function(optimal_weights) class="macro">#Portfolio variance portfolio_var = pd.DataFrame(columns=[&class="macro">#x27;Old Var&class="macro">#x27;,&class="macro">#x27;New Var&class="macro">#x27;],index=[class="num">0]) portfolio_var.iloc[class="num">0,class="num">0] = initial_portfolio_variance * (class="num">10.0 ** class="num">7) portfolio_var.iloc[class="num">0,class="num">1] = otpimal_variance * (class="num">10.0 ** class="num">7) portfolio_var.plot.bar() int_weights = (optimal_weights / optimal_weights[-class="num">1]) class=class="str">"cmt">// class="num">1 int_weights
「给EA接上权重与亏损触发线」
把 SciPy 算出的富时100组合最优权重写进 EA，是这一步的硬前提。下面这段代码里，权重数组长度是 11，对应 11 只成分股，数值从 -4 到 1 不等，负号代表该标的在组合里应偏空、正号偏多。 触发逻辑靠一个亏损限制参数：input double loss_limit = 20，意思是当账户浮亏超过 20 后，才可能调用最小化方差的程序去重新铺仓位。实际判断用的是 position_profit = 权益 - 余额，若这个值小于 -loss_limit，就进 minimize_variance()。 minimize_variance 函数会遍历股票列表，按 optimization_weights[i] 的符号和绝对值下单：大于 0 就循环买 0.3 手，小于 0 就循环卖 0.3 手，注释里标的是“FTSE Optimization”。注意原代码里 risk_minimized 初始置 true，导致后面的 for 循环根本不会执行，真要跑起来得先把这句改掉。 外汇与股指 CFD 杠杆高、回撤快，这类组合优化只降低方差、不消除爆仓可能，上 MT5 前先开模拟盘验证权重数组和触发阈值。
class="type">int optimization_weights[class="num">11] = {class="num">0,-class="num">2,class="num">0,class="num">0,-class="num">2,class="num">0,-class="num">4,class="num">0,class="num">1,class="num">0,class="num">1}; class="kw">input class="type">class="kw">double loss_limit = class="num">20; class=class="str">"cmt">// After how much loss should we optimize our portfolio? class=class="str">"cmt">//--- Should we optimize our portfolio variance using the optimal weights we have calculated if((loss_limit > class="num">0)) { class=class="str">"cmt">//--- Update the position profit position_profit = AccountInfoDouble(ACCOUNT_EQUITY) - AccountInfoDouble(ACCOUNT_BALANCE); class=class="str">"cmt">//--- Check if we have passed our profit target or if we are expecting a reversal if(((loss_limit * -class="num">1) < position_profit)) { minimize_variance(); } } class=class="str">"cmt">//+------------------------------------------------------------------+ class=class="str">"cmt">//| This function will minimize the variance of our portfolio | class=class="str">"cmt">//+------------------------------------------------------------------+ class="type">void minimize_variance(class="type">void) { risk_minimized = true; if(!risk_minimized) { for(class="type">int i = class="num">0; i < class="num">11; i++) { class="type">class="kw">string current_symbol = list_of_companies[i]; class=class="str">"cmt">//--- Add that stock to the portfolio to minimize our variance, buy if(optimization_weights[i] > class="num">0) { for(class="type">int i = class="num">0; i < optimization_weights[i]; i++) { Trade.Buy(class="num">0.3,current_symbol,ask,class="num">0,class="num">0,"FTSE Optimization"); } } class=class="str">"cmt">//--- Add that stock to the portfolio to minimize our variance, sell else if(optimization_weights[i] < class="num">0) { for(class="type">int i = class="num">0; i < optimization_weights[i]; i--) { Trade.Sell(class="num">0.3,current_symbol,bid,class="num">0,class="num">0,"FTSE Optimization"); } } } } }
◍ 一点提醒
这套把 Python 组合优化和 MQL5 动态加载打通的思路，核心价值在于让 EA 能跟着 MT5 里任意时间框架自己换参数，而不是写死在代码里。附带的 UK100.mq5 和 FTSE_100_AI.mq5 两个文件加起来不到 16 KB，真要验证，直接拖进终端就能跑。 后续若想同时压风险又盯无风险回报，只需改成本函数和约束，优化器换一个就行，底层框架不用动。外汇和贵金属杠杆高，这类自优化策略回测顺不代表实盘稳，先用模拟盘跑两周再谈真金白银。 作者留的 ZIP 里还有 369 KB 的笔记本，里面是最小方差组合的具体算例，有兴趣可以对照着把标的换成 XAUUSD 试试收敛速度。